Link Failure Recovery for MPLS Networks with Multicasting

Transcription

1 1/33 Link Failure Recovery for MPLS Networks with Multicasting Master s Thesis Presentation August 12 th, 2002 Yvan Pointurier Advisor: Jörg Liebeherr Multimedia Networks Group Department of Computer Science University of Virginia

2 2/33 Outline Introduction: MPLS, multicasting, tree repair problem A solution to the tree repair problem using graph theory MPLS multicast Fast Reroute mechanism Protocol design and implementation Experimental results Conclusion

3 3/33 Switching technology Circuit Switching PSTN Datagram Packet Switching IP Virtual Circuit Packet Switching ATM and MPLS

4 4/33 Recovery from link failures Traditional approaches to link failure recovery Physical layer Example: SONET ring wrapping after a failure Pro: fast (50 ms) Con: hardware solution high cost Network layer Example: rerouting after IP tables convergence Pro: software solution done by routing algorithms additional cost Con: slow convergence (several minutes) MPLS goals Recovery at the network layer Same recovery time as SONET using backup virtual circuits no

5 5/33 MPLS (Multiprotocol Label Switching) Provides Virtual Circuit Switching to IP networks Design started in 1997 by IETF, RFC 3031 was released in Ethernet header MPLS shim header IP header Data Label Identifier (20 bits) for a Virtual Circuit With Ethernet and IP: a label is contained in a shim header A label can be pushed, popped, swapped by MPLS routers Eth IP MPLS domain push L1 swap L1, L2 pop L2 Eth L1 IP Eth L2 IP Eth IP

6 6/33 MPLS: Signaling protocol Protocol responsible for virtual circuits management Two signaling protocols have been defined in MPLS: LDP/CR-LDP and RSVP-TE Example: establishing a label binding between two MPLS routers Upstream node for flow a U Direction for flow a Request for binding for a Label binding for a Downstream node for flow a D

7 7/33 MPLS Fast Reroute Name given to a rerouting technique with MPLS Pre-planned techniques are faster MPLS Fast Reroute provides a trade-off between cost and speed Objective: 50 ms rerouting time

8 8/33 MPLS Fast Reroute D E F A B C D MPLS Router Original path Backup path Notification path Link failure 6 steps: 1. Link failure detection 2. Link failure notification 3. Path switchover 4. Link recovery detection 5. Link recovery notification 6. Path switchback

9 9/33 Multicasting Goal: provide group communication B Shared multicast routing trees Multicast and MPLS Still in early draft stage A Core D C

10 10/33 The tree repair problem Pre-planned multicast tree protection against link failure Research Problem: Where should backup paths be placed? Objective: find an algorithm to place good backup paths in a multicast routing tree A A C C B B D D Unicast-style protection Cheaper protection with the same effect

11 11/33 In this Thesis, we Define a metric (resilience) which measures the degree to which a backup path protects a multicast tree Specify and analyze a Graph Algorithm for multicast tree protection against single link failure Propose protocol mechanisms ( MPLS multicast Fast Reroute ) for repair of multicast trees Design and implement the protocol mechanisms as extensions to MPLS signaling protocols Measure the performance of the implementation

12 12/33 Outline Introduction: MPLS, multicasting, tree repair problem A solution to the tree repair problem using graph theory MPLS multicast Fast Reroute mechanism Protocol design and implementation Experimental results Conclusion

13 13/33 A tree repair algorithm Given: Find: Network, multicast routing tree and a resilience metric Single backup path that maximizes the resilience of the multicast tree for single link failures Algorithm is presented as a graph algorithm: MPLS network graph Multicast routing tree tree embedded in the graph

14 14/33 Modeling process (1/2) A MPLS network E Link of a multicast tree Access link Other link Core C D MPLS Router B

15 15/33 Modeling process (2/2) A E Link of a multicast tree Other link of the network root C D Node attached to receivers B Other node of the tree Other node of the network

16 16/33 Link failure and backup path in a tree Link of a multicast tree Protected path Backup path root Node attached to receivers Other node of the tree

17 17/33 Formalization root A w=0 f=1 tdrop=2 w=0 f=1 tdrop=3 B E w=1 f=1 tdrop=1 w=1 f=1 tdrop=1 C D F G w l : f l : tdrop l : number of receivers downstream of a link l Link failure rate of link l number of receivers dropped when a link l fails tdrop l = w k k subtree l H

18 18/33 Defining Resilience for a tree with a backup path Resilience R(A,B) of a tree protected by a backup path between nodes A and B: R t = ( fk tdropk ) R (, ) = d A B ( f k tree R( A, B) = R R ( A, B) t k k tree k primary path( A, B) R t : number of receivers dropped on a link failure (no backup path) R d (A, B): number of receivers dropped on a link failure if a backup path is set between A and B R(A, B): number of receivers not dropped on a single link failure if a backup path is set between A and B d tdrop k )

19 19/33 Resilience of a tree protected by a backup path Large resilience R(A, B) more hosts are in the rearranged multicast tree on a single link failure We want to maximize the resilience This is equivalent to minimizing R d Find the two nodes of the tree A, B such that R ( A, B) min d = Y, Z nodes of the tree R d ( Y, Z)

20 20/33 Algorithm 1. Compute R d for all pairs of nodes It can be proven that one node of the pair that maximizes the resilience is a leaf R d can be computed in time log(n) with n=number of nodes in the tree 2. Choose the pair of nodes that minimizes R d 3. Compute the shortest path between the two nodes Time complexity: O(n 2 log(n)+(n-n) log(n-n)) =O(nNlog(N)) n = number of nodes in the tree N = number of nodes in the network Incremental version

21 21/33 Outline Introduction: MPLS, multicasting, tree repair problem A solution to the tree repair problem using graph theory MPLS multicast Fast Reroute mechanism Protocol design and implementation Experimental results Conclusion

22 22/33 MPLS multicast Fast Reroute (1/3) E C E C F D D G S S A J F G A H A multicast virtual circuit The mechanism consists of 6 steps B 1. Link failure detection Use probes Several missing probes failure 2. Link failure notification Propagate notification messages on the tree B

23 23/33 MPLS multicast Fast Reroute (2/3) S C D E F G A S C E H J B H J B A D F G 3. Switchover Traffic flows on the backup path The tree is repaired, no node is dropped

24 24/33 MPLS multicast Fast Reroute (3/3) C E C E S S D F G A J J H B A D F G H B 4. Link recovery detection Faster then link failure detection 5. Link recovery notification Same as link failure notification 6. Switchback Same topology as before the failure occurred

25 25/33 Outline Introduction: MPLS, multicasting, tree repair problem A solution to the tree repair problem using graph theory MPLS multicast Fast Reroute mechanism Protocol design and implementation Experimental results Conclusion

26 26/33 Design and implementation MPLS multicast MPLS multicast is not standardized Extension to MPLS-Linux unicast Duplicate packets inside the Linux kernel and forward the copies on different interfaces: mswap We provide a C API that enables users to create forwarding rules MulTreeLDP signaling protocol A label distribution protocol for MPLS multicast Support for Explicit Tree Routing Message formats are similar to those in LDP Implements MPLS multicast Fast Reroute

27 27/33 Hardware used for the experiments The Indra testbed 6 PC routers, PII-450, 128MB RAM, 4x100 Mbps Ethernet, Linux Fast enough to duplicate packets that arrive at full speed We achieved rates of 93Mbps even when each packet gets duplicated twice 1 sender 3 receivers

28 28/33 Testbed setup X PC3 PC5 PC1 PC2 PC4 PC6

29 29/33 Experimental results Link failure detection Two consecutive missing probes are interpreted as a link failure Distributed between 10 ms and 40 ms, average: 25.4 ms Link recovery detection First probe received after link failure is interpreted as link recovery Distributed between 0 ms and 10 ms, average: 5.5 ms Notification Measured 1.2 ms per hop Service interruption time Measured at PC5: 29.4 ms (average) PC1 PC2 X PC4 PC3 PC5 PC6

30 30/33 1. Experimental distribution of the service interruption times 14.0% 12.0% Mean: 29.4 ms Max: 49.6 ms Stdev: 7.1 ms 10.0% Frequency (%) 8.0% 6.0% 4.0% 2.0% 0.0% Repair time (ms)

31 31/33 2. Traffic received by PC ms Traffic received (MBytes) ms Time (s)

32 32/33 Conclusions Contributions Algorithm that selects a backup path MPLS multicast implementation for Linux Explicit Tree routing MPLS multicast Fast Reroute Design, implementation and evaluation Future work Algorithm: multiple link/node failure - testing Standardize MulTreeLDP Avoid packet duplicates on switchback Implementation in larger networks

33 33/33 Questions and Answers Thanks for your attention! After the Questions and Answers session, I will show a demo of MPLS multicast Fast Reroute in the MNG lab